− 1 − Basel and Procyclicality: A comparison of the Standardised and IRB Approaches to an Improved Credit Risk Method By C. Goodhart and M.Segoviano 1. Introduction. Our procedure here is to try to reconstruct a typical bank portfolio for a country and then, holding the presumed loan book unchanged over time, (i.e. replacing failed loans with loans of a similar quality), to examine how the loan ratings would have shifted, and hence how the capital adequacy requirements (CAR’s) for the banks would have varied over time; for other similar exercises see Kashyap and Stein (2003 and 2004) and Gordy and Howells (2004). To do this we use Moody’s data on U.S. corporate bonds, included on Moody’s Investors Service, Credit Risk Calculator. We can only do this exercise for those countries for which Moody’s data on credit ratings has a long enough time series. Unfortunately this rules out most large European countries since adequate Moody’s data only go back to 1988 for the U.K., 2001 for Germany; 2002 for France; 2003 for Italy; 2002 for Spain. In practice we also used data provided by the Mexican Financial Regulatory Agency and the Norwegian Central Bank on Corporate Loans for these latter two countries. The Mexican data incorporates statistics between 1995 and 2000 and the Norwegian data incorporates statistics between 1988 and 2001. For an earlier exercise along these same lines, and using the same Mexican data set, see Segoviano and Lowe, (2002). Amongst the problems are how to reconstruct a `typical’ bank portfolio; whether, and how, to deal with the problem of failing loans dropping out of the portfolio; and what account to take of the fact that Basel II is a regime change that may make banks alter their `typical’ behaviour. Very briefly, we reconstructed a typical bank portfolio as follows. We assumed that each portfolio consisted of 1000 loans, each one with equal exposure. From each country’s data sources, we obtained the through time proportion of assets (bonds for the U.S. or corporate loans for Mexico and Norway) that were classified under each of the
Transcript
− 1 −
Basel and Procyclicality: A comparison of the Standardised and IRB Approaches to an
Improved Credit Risk Method
By C. Goodhart and M.Segoviano
1. Introduction.
Our procedure here is to try to reconstruct a typical bank portfolio for a country and
then, holding the presumed loan book unchanged over time, (i.e. replacing failed
loans with loans of a similar quality), to examine how the loan ratings would have
shifted, and hence how the capital adequacy requirements (CAR’s) for the banks
would have varied over time; for other similar exercises see Kashyap and Stein
(2003 and 2004) and Gordy and Howells (2004). To do this we use Moody’s data on
U.S. corporate bonds, included on Moody’s Investors Service, Credit Risk Calculator.
We can only do this exercise for those countries for which Moody’s data on credit
ratings has a long enough time series. Unfortunately this rules out most large
European countries since adequate Moody’s data only go back to 1988 for the U.K.,
2001 for Germany; 2002 for France; 2003 for Italy; 2002 for Spain. In practice we
also used data provided by the Mexican Financial Regulatory Agency and the
Norwegian Central Bank on Corporate Loans for these latter two countries. The
Mexican data incorporates statistics between 1995 and 2000 and the Norwegian data
incorporates statistics between 1988 and 2001.
For an earlier exercise along these same lines, and using the same Mexican data
set, see Segoviano and Lowe, (2002). Amongst the problems are how to reconstruct
a `typical’ bank portfolio; whether, and how, to deal with the problem of failing loans
dropping out of the portfolio; and what account to take of the fact that Basel II is a
regime change that may make banks alter their `typical’ behaviour. Very briefly, we
reconstructed a typical bank portfolio as follows. We assumed that each portfolio
consisted of 1000 loans, each one with equal exposure. From each country’s data
sources, we obtained the through time proportion of assets (bonds for the U.S. or
corporate loans for Mexico and Norway) that were classified under each of the
− 2 −
reported ratings for a given country. With this information we constructed the
benchmark portfolio that we used to compute capital requirements at each point in
time.
By assuming that the initial bank loan book remains unchanged throughout, this is
equivalent to assuming that failed loans are replaced by loans of similar initial quality.
This is what Kashyap and Stein (2004) did, and seems natural. Gordy and Howells
(2004) argue, however, that banks will aim for a higher quality portfolio during
recessions, and thus will replace failing loans with credits of higher, than initial,
quality. At the macro level it is hard, in most countries, to see where the supply of
such higher quality loans would come from during recessions; in discussion of this
point at a BIS Conference in May 2004, Michael Gordy noted that in the USA high
quality companies tended to shift their borrowing from capital markets, e.g. the
commercial paper market, to banks during recessions. In any case, since risk
spreads on lower quality loans widen during recessions, any extra benefit to the bank
would be slight. So we feel relatively comfortable about our own assumption.
The results of this exercise for the three countries examined are stark. We compared
the implied capital requirements for our `typical’ bank under three regulatory regimes;
first the standardised approach in Basel II, (which is close to that applied in Basel I);
second, the Foundations IRB approach, (i.e. assuming a constant Loss Given
Default, since we have no good time series in any country for average LGD); and
third, an Improved Credit Risk Method (ICRM). This latter uses a Merton approach to
model credit quality changes and an indirect approach to model correlations amongst
the individual credits in the overall portfolio. The construction of an ICRM is, however,
quite complex. The main addition in this note, beyond our associated work, available
at Goodhart, Hofmann and Segoviano (2004), is that we spell out in more detail here
how we do the exercise of estimating the required capital requirements under an
Improved Credit Risk Method (ICRM).
The outline of the paper is as follows. In section 2, we elaborate on the need to
measure portfolio effects for proper credit risk quantification. In section 3, we develop
the ICRM. In section 4, we present the empirical implementation and results. This
section also describes the data and assumptions made to perform the exercise.
Finally, our conclusions are summarised in section 5.
− 3 −
2. Why a portfolio Approach? Since the quality of the credit portfolio of a bank can change at any time in the future,
there is a need to make frequent calculations of the expected losses that a bank
could suffer under different risk situations. This analysis of uncertainty is the essence
of risk management. Therefore, measuring the uncertainty or variability of loss and
the related likelihood of the possible levels of unexpected losses in a bank’s portfolio
is fundamental to the effective management of credit risk. Sufficient earnings should
be generated through adequate pricing and provisioning to absorb any expected
loss. However, economic capital should be available to cover unexpected credit
default losses, because the actual level of credit losses suffered in anyone period
could be significantly higher than the expected level. The estimation of the profit and
loss distribution of credit portfolios, from which the unexpected losses can be
identified (e.g. 99.9 Percentile loss level), represents the issue to be addressed in
this document.
Figure 1: Credit Portfolio Profit and Loss Distribution (P&L).
Portfolio Losses
The adoption of the portfolio approach to risk analysis (Markowitz (1959)) has been
amply documented and adopted in numerous finance applications1. Under this
theory, investors seek an optimal risk-return relationship when formulating their
investment portfolio.
1 See Cochrane (2001) for numerous examples on asset pricing.
− 4 −
The model presented in this paper provides a methodology for assessing portfolio
risk due to changes in loan value caused by changes in obligor (ie. borrower’s) credit
quality. Changes in value are caused not only by possible defaults, but also by
upgrades and downgrades in credit quality; the correlation of credit quality variations
across obligors in the portfolio is also considered. This allows us to calculate the
benefits of diversification in the portfolio. Credit risk modellers have already
developed risk management techniques that seek to take account of this portfolio
diversification effect2. In this paper we present a modification to the “Credit-Metrics”
and KMV methodologies that have been used to simulate unexpected losses from
credit risk in analysed portfolios. For detailed exposition refer to the Credit-Metrics
and KMV technical documents. We refer to our modification to the “Credit-Metrics”
and KMV methodologies as an Improved Credit Risk Model: ICRM3.
3. An Improved Credit Risk Method (ICRM).
As already stated, our model assesses portfolio risk arising from changes in loan
value caused by changes in obligor credit quality. Given the composition of a
particular portfolio, all the possible portfolio values and their probabilities are
recorded in the profit and loss (P&L) distribution of the portfolio. This distribution
records both, increases and decreases in the value of the portfolio caused by the
upgrades and downgrades in the loans’ credit qualities. The modelling of the Profit
and Loss distribution of the portfolio (P&L) can be broken down into the following
steps:
3.1 Modeling the distribution of a single loan.
3.1.1 Credit risk migration and the Merton approach.
3.1.2 Loan valuation.
3.2 Portfolio risk calculation.
3.2.1 Joint probabilities.
3.2.2 Indirect approach to model correlations.
3.2.3 Simulation of quality scenarios for the credit portfolio.
3.2.3 Valuation, P&L distribution and unexpected losses.
2Such approaches may be subject to further improvements, but it is not our intention have to suggest possible improvements to each methodology. 3 The term “Improved” refers to the fact that this model does take account of the benefits of diversification. This is in contrast to the IRB approaches that use a “simplified, single risk factor model” See Secretariat of the Basel Committee on Banking Supervision, (2001).
− 5 −
3.1 Modelling the distribution of a single loan. 3.1.1 Credit Risk Migration and the Merton Approach.
The Merton approach assumes that a firm’s equity can be viewed as a call option on
the firm’s assets with a strike price equal to the book value of the firm’s debts (Merton
(1974)). The intuition behind this assumption is that, given the limited liability of
equity, equity holders have the right, but not the obligation to payoff debt-holders and
take over the remaining assets of the firm. This approach implies that the credit
quality (rating) of a given debtor is related to the difference between the market value
of its assets and its debt.
Under this approach, the change in the value of the assets of a given company is
related to the change in its rating. So, the distribution of the company’s asset returns
can be used to calculate the distribution of the probability of the firm’s rating change.
For the generalisation of this model, it is necessary to include, in addition to the
default state, different credit quality states. This is because in this model, risk comes
not only from default but also from changes in value due to up(down) grades.
Figure2: The Distribution of Assets’ Returns.
AE CD B
Z e Z d Z c Z b
The likelihood of any credit rating migration in the coming period is conditioned on
the current credit rating of the obligor.
Individual likelihoods of migration are usually represented in matrix form. This table is
called a transition matrix. The transition matrix is the table that summarises the
migration probabilities from one credit quality to any other.
− 6 −
Table1: Transition Matrix.
A B C D E A 0.865047 0.054462 0.011523 0.002826 0.002439B 0.057558 0.806042 0.106912 0.006859 0.014964C 0.005934 0.052618 0.812763 0.060135 0.065386D 0.001516 0.009098 0.058378 0.708470 0.222538E 0.000000 0.000000 0.000000 0.000001 0.999999
To read this table, the credit rating (today) at time t0 is written on the extreme left column. The possible
ratings to which a given loan can migrate at the risk horizon, t1 are written on the top row. For example,
a loan that at t0 is rated as C has 81.2763% probability of remaining in the same rating at t1. The table
indicates that there is a 6.0135% probability that the loan will migrate to a D rating at t1 and there is a
6.5386% probability that the loan will default (column E) at t1. The transition matrix also indicates that
there is a 5.2618% that the loan will migrate to a B rating and so on.
Having the transition probabilities between different credit qualities, and employing
the Merton Approach, it is possible to derive the (market) value of assets that
represent the cut-off values between different credit qualities, as shown in Figure 1.
These cut-off values fulfil the condition that if the change in the market value of the
asset (r) is sufficiently negative, (i.e. smaller than Ze), then the credit falls into
default; if Ze < r < Zd, the credit is rated D, and so on. Taking into consideration the
empirical transition matrix, the cut-off values are obtained by solving the following
equations (e.g. for a loan initially rated as X):
Prob(E|X) = Prob(r < Ze) = ϕ(Ze) (1)
Prob(D|X) = Prob(Ze <r < Zd) = ϕ(Zd) - ϕ(Ze)
Prob(C|X) = Prob(Zd <r < Zc) = ϕ( Zc) - ϕ(Zd)
Prob(B|X) = Prob(Zc <r < Zb) = ϕ(Zb) - ϕ(Zc)
Prob(A|X) = Prob(Zb <r < Za) = 1 - ϕ(Zb)
Where, R is the implied market value of assets, and ϕ is the Normal Cumulative
Density Function (CDF).
3.1.2 Valuation.
In the previous section, we determined the likelihoods of migration to any of the
possible credit quality states at a given risk horizon. In this section, the values at the
risk horizon for these credit quality states are determined. Values are calculated for
− 7 −
each migration state. These valuations fall into two categories. First, in the event of
up(down) grades, only the change in the value of the bond due to migration is
considered. To obtain the values at the risk horizon corresponding to rating
up(down)grades, a straightforward present value re-valuation is performed. This
revaluation upon up(down)grade accounts for the decreasing likelihood that the full
amount of the loan will be repaid as the obligor undergoes rating downgrades, and
the increasing likelihood of repayment if the obligor is upgraded. Second, in the event
of default, the change in the value of the loan due to its downgrade (to the default
category) is estimated in the same manner; however, the remaining value of the loan
is multiplied by its loss given default (LGD)4.
Table2: Valuation Table.
A B C D E A 0.000000 -0.012525 -0.062947 -0.220099 -0.997560B 0.012525 0.000000 -0.050422 -0.207575 -0.985035C 0.062947 0.050422 0.000000 -0.157152 -0.934613D 0.220099 0.207575 0.157152 0.000000 -0.777461E 0.997560 0.985035 0.934613 0.777461 0.000000 To read this table, the credit rating (today) at time t0 is written on the extreme left column. The possible
ratings to which a given loan can migrate at the risk horizon, t1, are written on the top row. Changes in
the value of the loan due to migration are in the body of the table. For example, if a loan that at t0 is
rated as C, remains at the same rating at t1, has a zero present value change. If the same loan migrated
to a D rating, its present value would be decreased 15.71%. If the loan were upgraded to a B rating, its
present value would be increased 5.04%, and so on. This re-valuation upon downgrades/upgrades
accounts for the decreasing/increasing likelihood that the full amount of the loan will be repaid as the
obligor undergoes migrations.
As already stated, given a current credit rating of the obligor the likelihood of any
credit rating migration in the coming period is conditioned on the current credit rating
of the obligor.
With the transition probabilities indicated by the transition matrix and the possible
values within each migration state indicated by the valuation table, it is possible to
obtain the value distribution for each exposure on a stand-alone-basis. Beyond this,
portfolio credit risk models5 then extend this framework to the portfolio as a whole, in
order to obtain the distribution of value of the complete portfolio, the so called profit 4 The loss given default is estimated as LGD= 1-(percentage of recovery value). When databases allow it, the recovery rates and consequently, the LGD’s are estimated based on loan characteristics, e.g. credit quality of debtor, geographical area, etc. 5 See CreditMetrics (1997) and CreditRisk+ (1997) technical documents for specific details.
− 8 −
and loss distribution (P&L) from which we will derive the Value at Risk (VaR) figure
used to define capital requirements.
3.2 Portfolio risk calculation.
In section I, we explained the steps followed to obtain the credit risk for a stand-alone
exposure. Here we extend the methodology to a “portfolio”. For reasons of parsimony
the methodology explained here will refer to a portfolio of just two exposures;
however, the methodology applies to a portfolio of any number of elements. In
general, the necessary steps are the same as in the previous section, but there is
one significant addition. Now it becomes necessary to include the contribution to risk
brought about by the effects of credit quality correlations. So, first, we will discuss the
joint likelihoods of credit quality co-movements. Second, we extend the credit risk
calculation for the stand-alone exposure to the multiple exposure case.
3.2.1 Joint likelihood in credit quality. Understanding joint likelihoods allow us to account properly for portfolio
diversification effects. Correlations determine how often losses occur in multiple
exposures at the same time. The portfolio value volatility (risk) will be lower if
correlations between credit events are lower.
In theory, a correlation matrix of changes of credit quality between creditors can be
computed by developing an explanatory model of the changes in the value of the
assets of the borrowers. This approach presents several practical problems for
implementation, the most important being the handling of very large correlation
matrices. Additionally, it is not possible to obtain the changes in the market value of
assets for each particular borrower, since it would be necessary to have specific
information about the internal financial structure of each borrower. These two
disadvantages make it impossible to implement an ideal correlation matrix; for these
reasons we will adopt an indirect (but more manageable) method to introduce the
portfolio diversification effect.
− 9 −
3.2.2 Indirect approach to model Correlations. The referred indirect method for introducing the portfolio diversification effect was first
presented in Segoviano (1998). It is based on an assumption made by the
CreditMetrics methodology. This methodology makes an a-priori distinction of the
factors that determine the changes in the value of the assets of the borrowers. This
distinction comes from two basic components: the market component and the
idiosyncratic component. By definition, the idiosyncratic component does not
correlate with anything, since it refers to those factors unique for the debtor. But the
market component can then be further disaggregated into several separate
components that allow the portfolio diversification.
rtotal = WM rmarket + WI rIdiosyncratic (2)
Where:
WM: Percentage of returns explained by the market component6.
rmarket: Market component of returns. WI : Percentage of returns explained by the idiosyncratic component.
Iidiosyncratic: Idiosyncratic component of returns.
Next, the market component of Returns can be defined as:
HA: Percentage of the market component explained by the GDP returns of the
borrowers’ country (geographical location). The determination of HA will be explained
in Section 3.
rGDPGeographicalLocation: Borrower’s country (geographical location) GDP’s return. (1-HA): Percentage of market component explained by the GDP returns’ of the
6 In the CreditMetrics technical document, how these weights can be calculated is explained. After empirical implementations, an acceptable value of WM = 70% is derived. For our exercise, we also assume this value.
− 10 −
Once all the elements that compose the market component of assets’ returns have
been considered, the next step is to calculate the correlations between the
borrowers’ loans making up a credit portfolio.
Given a pair of borrowers, classified under ratings X and Y, whose sectoral activities
are B and V; located in A and E country groups, and with returns expressed in the
following way:
r w r w H r w H rX IX IX MX A A MX A B= + + −( )1
r w r w H r w H rY IY IY MY E E MY E V= + + −( )1
The problem of estimating the correlations among each couple of borrowers in the
portfolio is summarised in the following way:
BVEMYAMXAEEMYAMXXY HwHwHwHw ρρρ )1()1( −−+= (4)
Where:
pAE: is the correlation between different country groups.
pBV: is the correlation between different sectoral activities.
This equation is computed for each pair of borrowers making up the portfolio. The
results of computing this equation are compiled in a (n x n) square matrix, where n is
the number of loans in the portfolio. This matrix is named the correlation matrix
between borrowers and is unique for each portfolio. This matrix is a key variable for
the simulation of unexpected losses, since it incorporates the necessary elements to
quantify the concentration/diversification of the portfolio.
As explained above, the transition matrix indicates the probabilities of quality
changes that a stand-alone exposure with a given rating might experience.
Additionally, when correlations of quality changes between borrowers are involved,
we can compute the joint likelihood of credit up(down)grades between the loans
making up a portfolio.
Debtors with similar characteristics will tend to migrate jointly to different credit
qualities when hit by economic shocks. Debtors with different characteristics will tend
− 11 −
to migrate separately to different credit qualities when hit by economic shocks. This
implies that credit portfolios concentrated in credits with similar characteristics will
tend to have higher unexpected losses since they will not be diversifying the possible
economic risks.
With these components, we show in the following section how quality scenarios for
the portfolio are simulated. From these quality scenarios, the loss distribution is built
from which it is possible to obtain an estimate of unexpected losses.
3.2.3 Simulation of Quality Scenarios for the Credit Portfolio.
Combining the transition matrix with the correlation matrix between borrowers, and
under the Merton framework that assumes lognormal asset returns (see equation
(1)), we obtain the joint likelihood of credit quality migration and simulate credit
quality scenarios. The simulated quality changes of the components of the portfolio
allow us to estimate the losses or profits that determine the P&L distribution of the
portfolio.
In order to generate these scenarios, the following process is undertaken:
1. Generation of random uniform numbers.
2. Transformation of such random numbers into normal standard random
numbers.
3. Transformation of the normal standard random numbers into normal-
multivariated random numbers with a correlation matrix defined by the
correlations between creditors.
3.2.4 Valuation, P&L Distribution and Unexpected Losses.
Once the credit portfolio quality scenarios have been simulated, we use the
simulated credit quality scenarios to re-evaluate the portfolio exposures as explained
in section I.2. With the portfolio exposures re-evaluated, we obtain the P&L
distribution for the portfolio.
This is done by computing the losses/gains that come from the difference between
initial and final credit qualities in the loans making up the portfolio. The losses/gains
− 12 −
obtained from the simulation process are used to build a histogram. This histogram
summarizes the loss distribution of the credit portfolio.
From this distribution a Value at Risk (VaR) is defined from which we obtain the
amount of unexpected losses from the portfolio. The unexpected losses divided by
the total amount of the portfolio represent the percentage that with a given probability
(defined by the chosen percentile) could be lost in an extreme event. Capital
requirements should be such that they can cover these losses.
4. Empirical Implementation and Results.
Our objective here is to try to reconstruct a typical bank portfolio for a country and
then, holding the presumed loan book unchanged over time, (i.e. replacing failed
loans with loans of a similar quality), to examine how the capital adequacy
requirements (CAR’s) for the banks would have varied over time. We have assumed
that each portfolio consisted of 1000 loans, each one with equal exposure. Below we
explained the assumptions taken for the additional variables that were needed to
perform this exercise. We also indicate the databases from which the necessary
information was taken.
− 13 −
4.1 Data and Assumptions. Geographical distribution of exposures: From the BIS database on consolidated banking claims for the U.S. and Norway, we
obtained the through time proportion of assets invested in different geographical
areas7. Information for Mexico was obtained from the databases provided by the
Mexican Financial Regulatory Agency (CNBV).
Credit quality distribution of exposures: We obtained the through time proportion of corporate bonds that were classified
under each of the reported ratings for the U.S. from the Moody’s investors service
database. In the case of Mexico and Norway, we obtained the through time
proportion of corporate bonds that were classified under each of the reported ratings
from the databases provided by the Mexican Financial Regulatory Agency and the
Norwegian Central Bank.
Sectoral Activity distribution of exposures: We assumed that the simulated portfolios consisted of loans evenly distributed
across the major sectoral activities that comprise GDP8.
Transition matrices: We use Moody’s data on U.S. corporate bonds, included on Moody’s Investors
Service, Credit Risk Calculator. In the event we also used data provided by the
Mexican Financial Regulatory Agency and the Norwegian Central Bank on Corporate
Loans. The Mexican data incorporates information between 1995 to 2000, and the
Norwegian data incorporates information between 1988 to 2001.
Loss Given Default:
We fixed this at 50% in order to make results comparable to the IRB foundation
approach developed by the Basel Committee.
7 Developing: Africa and the Middle East; Asia and Pacific; developing Europe; Latin America. Developed: EU (non-EMU); EMU; Other Industrial; offshore centres. 8 We included the following sectoral activities: financial, building, mining, information technology, retail, textile, chemical, energy, pharmaceutical, tobacco, food production, beverages, electrical.
− 14 −
Market component of returns:
In equation (3) we assume that firms’ market component of returns are explained by
both the firms’ sectoral activities and geographical locations. In order to get a proxy
of the percentage of the market component of returns that is explained by
geographical location (1-HA), we run the following OLS regressions:
rMarket = C+B rGDPGeographicalLocation + є (5)
Where:
C: is a drift term
rMarket: was obtained by estimating the returns of the Morgan Stanley Capital International
(MSCI) indexes for major sectoral activities9.
rGDPGeographicalLocation: was obtained by estimating GDP growth rates of the analysed
countries.
In general, in regression analysis, the percentage of the total variation of a
dependent variable that is explained by the assumed explanatory variables is
indicated by the measure of goodness of fit, R2 (explained sum of squares over total
sum of squares). Therefore, we took the R2 that was obtained by running the
regressions specified in equation (5) as proxies for the percentage of market returns
that is explained by the GDP growth rates of the analysed countries, e.g., we take
R2~(1-HA). Consequently, (1- R2) ~HA.
Correlations among different country groups and economic activities:
In equation (4), we make use of AEρ , the correlation between different country
groups and BVρ , the correlation between different sectoral activities. The first were
computed using the spreads of syndicated loans for each country group. We
assumed that such spreads measure the riskiness of the financial system in each
9 These indexes are composed as weighted averages of prices of the major corporates in developed economies for specific sectorial activities. The sectorial activities that we considered were: financial, building, mining, information technology, retail, textile, chemical, energy, pharmaceutical, tobacco, food production, beverages, electrical. Source: Datastream.
− 15 −
country group. The latter were computed from indices of the major sectorial activities
examined in the exercise10.
We used the variables and assumptions described in this section to perform the
simulation of credit quality scenarios with which we re-evaluated the exposures in the
portfolio and computed the P&L of the portfolio.
Simulation:
In this application, we programmed an algorithm to compute 10,000 possible quality
scenarios for each of the (n x n) couples of the loans that make up the portfolio. Each
quality scenario shows a change in the market value of the borrowers’ assets whose
loans compose the portfolio. Since it was assumed that the process that generates
changes in the assets’ log-returns follow a normal distribution, we use a normal-
multivariated distribution to generate joint quality migrations.
10 Idem footnote 9.
− 16 −
4.2 Results. The results of this exercise for the three countries examined are stark. We have
simulated the time paths of CARs under each of our three approaches, standardised,
IRB Foundation (IRB F) and ICRM, for our various countries, and the results are set
